This paper presents an artificial intelligence (AI) architecture for simulating the iterative updating of the human working memory system. This architecture features multiple interconnected neural networks designed to emulate specialized modules of the cerebral cortex. These networks are hierarchically organized, integrated into the global workspace, and can temporarily maintain high-level representational patterns similar to the psychological items held in working memory. This maintenance is achieved through persistent neural activity (both through persistent neural firing and synaptic strengthening). These persistently maintained representations are repeatedly replaced, resulting in a gradual change in the content of the working memory system. As the content gradually evolves, successive processing states overlap and are continuously interconnected. In this paper, we explore how this architecture drives iterative changes in the distribution of co-active representations, ultimately enabling mental continuity between processing states and, ultimately, human-like thought and cognition. Like the human brain, this AI working memory store is connected to multiple image (topological map) generation systems corresponding to different sensory modalities. As working memory is repeatedly updated, the maps generated in response constitute a sequence of related mental images. Therefore, neural networks that emulate the interactions between the prefrontal cortex and the early sensory and motor cortices capture the visual guidance capabilities of the human brain. These sensory and motor image generation, combined with a repeatedly updated working memory store, could provide AI systems with the cognitive assets necessary to achieve artificial consciousness or artificial emotions.
